GenAI in Payroll: Fact, Fiction, and The Future with Special Guest Sheri Sullivan

[00:00:00] Welcome everyone to another episode of HR amp Payroll 2.0. I'm Pete Tilly-Akkus and as always

[00:00:13] I'm joined by the legendary Julie Fernandez. Welcome Julie. Thanks Pete. I great to be here with you today

[00:00:18] and brought along a special guest. Yes, I'm so excited. I love when you bring company it's always exciting.

[00:00:24] Cool why don't we get right to it and introduce Sherry. Let's do it. So today we're going to be talking

[00:00:28] with Sherry Sullivan who's a partner in global payroll operate leader at Ernst & Young. I think I've

[00:00:34] known her for a couple of decades maybe at least and always in this payroll, global payroll space

[00:00:39] so I'm super excited to bring her into the conversation and Sherry welcome. Thank you so much Julie

[00:00:48] and Pete it's a pleasure to be here. Yeah welcome could you have you Sherry excited to talk to you

[00:00:53] about AI right? Yes definitely something if you would have asked me a couple of years ago

[00:00:59] that I would be talking about I would tell you know but certainly our whole life has changed.

[00:01:04] Yeah 100% I'm a practitioner of payroll myself from from early early days and I think about

[00:01:08] that sometimes too like the things that we are we're having at our disposal and the things

[00:01:14] that are proliferating it's it's truly futuristic and it's an exciting time to be in payroll for sure.

[00:01:19] Absolutely well hey Sherry so we have this we have this question we start with when introducing

[00:01:26] our guests and we ask everyone how they got started in payroll or like what was their entry point

[00:01:32] or how did they get started and then after that we want to know what keeps you here. Yeah.

[00:01:38] Well great well I think none of us raise our hand and say when I grow up I want to be in payroll

[00:01:42] right yeah certainly so how I got started in payroll is I had a gentleman call

[00:01:49] and asked me to come over and actually look at his business and help him to figure out

[00:01:54] how to take it internationally and that was actually the CEO and founder and now chairman of the

[00:02:00] board of paychecks Tom Gowasano and so I thought it was an interesting enough model I went to go

[00:02:05] meet with him and quite frankly I thought oh this will be I was a consultant at the time this

[00:02:10] will be an interesting two to three year project and I'll do have some fun along the way but as you

[00:02:16] can see I got hooked on it I stayed on it I've been in payroll now for 20 almost 23 years

[00:02:23] believe it or not and I think what keeps me in payroll is the fact that there's still so much left

[00:02:30] to do I think the models have lagged behind obviously the technologies lagged behind and it's such

[00:02:37] the heartbeat of every single employer every single you know employee obviously we all love what we

[00:02:44] do but let's be honest if we weren't getting paid would we still do it so I think that's what keeps

[00:02:48] me as the whole challenge and the fact that you know that no one's really solved it I think the only

[00:02:53] other thing I've had is I think and I would be interested in your thoughts but I've seen payroll

[00:02:58] change more over the last two to three years than it had in the prior 20 100 percent I would

[00:03:04] absolutely agree I sit and look at some of the things that are coming about and the way firms

[00:03:09] are deploying it and just the cool use cases they're applying it to and it's like where was

[00:03:12] that when I was a practitioner right managing huge payrolls and dealing with you know all that

[00:03:18] comes with that it's such a blessing to practitioners and it's elevating this rich skill set they have

[00:03:24] and exposing that I think for the for the good right giving people a lot of great opportunities so

[00:03:28] good good that you made it and good that you're still here awesome oh my gosh I'm gonna have

[00:03:35] some dogs yeah okay so why don't we just jump in a little bit in yeah and kick off the whole topic

[00:03:42] of AI which is maybe like the biggest hot topic on the planet this year for sure everybody's talking

[00:03:48] about it and I'm really excited that we could focus on AI and payroll and that combination together

[00:03:54] with you Sherry and maybe one of the first things to just talk about is when folks say AI like

[00:04:01] are we talking about AI artificial intelligence are we talking just about automation that makes

[00:04:05] business processes easier and simpler you know like what do you see as the differences among all

[00:04:11] these terminologies and these things that get tossed out and labeled AI when in today's language yeah

[00:04:20] that's a great question because Sherry I think just to add something real quick I think a lot of

[00:04:23] folks sort of lump automation into a lot of right don't lump it into one bucket but there's

[00:04:28] such a different diversity in what we're talking about right yes there absolutely is I mean I think

[00:04:33] you know obviously we've all learned or I certainly have thanks to all of the experts here to

[00:04:39] be why and being part of a lot of those processes that artificial intelligence has been around for years

[00:04:46] and years and certainly we've seen it migrate from having you know some abilities that need to

[00:04:53] be programming in we've obviously seen the machine learning we've seen some deep learning but

[00:04:58] we really are in a place of gender to AI and so I think your questions really great because I

[00:05:03] think we do tend to lump everything into that one category and make generalizations but they're

[00:05:08] completely different and I think what we've seen over the last 18 or 19 months is as a quantum

[00:05:16] shift is a quantum shift in what is gender to AI and gender to AI I think how is it different from

[00:05:22] some of these other whether it's RPAs or other processes you know gender to AI is the type of

[00:05:29] artificial intelligence that can create new content I think that's pretty important because that

[00:05:34] means it can create images it can create text it can create audio or video and it actually can be

[00:05:41] trained in the you know the other ways and the other forms of automation and artificial intelligence

[00:05:48] and machine learning you had to really give it you know A you know A is the question B is the answer

[00:05:54] these go together it's you had to really be very very prescriptive but now with our large language

[00:06:00] models and I've personally seen this with a bunch of the P.O.C.'s and client pilots that we've done

[00:06:06] it actually can take information regardless of language and be able to actually synthesize

[00:06:13] it be able to answer questions based on it and that is why we're we're in a whole different realm

[00:06:20] an area when it comes to gender to AI and that really is the definition of gender to AI is

[00:06:25] an artificial intelligence able to learn and grow now I can tell you one of the things that's

[00:06:30] really interesting about that is it also you can't necessarily predict it right so some of the

[00:06:35] concerns is how do I control it and and really put up those rails yeah imagine you know like there's

[00:06:40] there's a scary element to that too right like well if it's if it's thinking and pushing through

[00:06:45] and making sense of certain things how will it actually string them together and you know

[00:06:50] and will I be okay with what it puts together and what that output looks like so I'm imagining

[00:06:56] there's an awful lot of challenges around controls or even being able to you know kind of monitor

[00:07:02] have some parameters around the outcomes yeah I think that I think the exciting thing for me

[00:07:07] to emphasize in generative is that contextual understanding and the learning as you mentioned

[00:07:12] that that progresses where it learns and comes back with better understanding the next time right

[00:07:17] in the next time and so when you think about all of the help that that can provide to the practitioner

[00:07:22] to augment to the to the employee to understand the check to leaders wanting to understand reports

[00:07:27] or receive guidance from those reports I think it's all just there's such infinite impact that can

[00:07:33] happen for for HR and payroll it's super super exciting really yes really sure you the skeptic

[00:07:39] in me always says you know like is it really possible and payroll and is it really generating

[00:07:46] and and and amalgamating things together in payroll that isn't just returning a number or figure

[00:07:52] so I'm so curious to hear a little bit more from your perspective around what is actually really

[00:07:58] happening in payroll AI wise today and what and what you're seeing I can do and where you're

[00:08:05] seeing that going yeah so so maybe if we think about some of the use cases and I'll get into

[00:08:11] some specifics if you'd like as to how we're using it or how we've seen it used with our clients

[00:08:17] and and I think the first part is you know Kennet and this was probably the first use case that

[00:08:23] you may have seen that we we kind of started on is how can it can it be used or the hypothesis was

[00:08:31] Kennet the use to actually answer employee questions for tier zero and tier one more completely

[00:08:37] more accurately um and and the first step was Kennet do it with generic and general information

[00:08:46] and the second was Kennet do it with PII data and I think one of the things that was so

[00:08:54] you know almost crazy for us to experience is that it was able a given example before with artificial

[00:09:01] intelligence or RPA or anything you had to actually spend a lot of time programming it

[00:09:06] and in order to program it you actually had to understand the and be trained on the technology

[00:09:12] and on how to program whether it was zero in one's or whether it was some kind of Python script

[00:09:18] or whether it was an RPA script etc but this technology now allows practitioners and business

[00:09:26] experts to actually weigh in and be part of that what is called you know in prompt engineering for

[00:09:32] example and that means that you can actually just with a simple thumbs up thumbs down and then some

[00:09:38] correction on the back end around those some of the prompt engineering actually change the way

[00:09:43] it answers um and it does really learn through those processes so so the first use case was this

[00:09:50] Kennet answer employee questions and I can tell you three of the things that we found out one is

[00:09:55] it didn't matter what source the data what language the data was in so one of the things we did

[00:10:02] was we put in a ton of information about regulatory compliance just to test it and we put it in

[00:10:08] in I don't know a hundred countries but by the way the source document was all English

[00:10:13] and then we had people around the world and you know 50 languages plus try and break it

[00:10:20] try and ask questions in their language about their country and see if it could actually do it

[00:10:25] and one of the things that was really interesting is the capabilities right now already have it

[00:10:29] and we've all been through translations that were the words were correct but the meaning didn't

[00:10:35] mean anything but this is totally different so that was just one example of the fact it's able

[00:10:40] by the way it absorbed masses amount of data in very little to any time and within five days of

[00:10:48] just the team working on some prompt engineering it was able to actually increase the accuracy

[00:10:53] of answers to over 95 percent which is way more than the human we did some some benchmarks on

[00:11:00] human answers if humans answers were only able to get in the high 60s so so that's the kind of

[00:11:07] impact we're seeing does that make sense? So some of these some of these some of the early use

[00:11:12] cases that you started talking about were really on the inquiry support side right and that's right

[00:11:17] yeah and then I just want to clarify think what I'm hearing and you know what maybe I wanted to

[00:11:25] make a point of is this is this is AI that is contextualizing and ingesting information

[00:11:33] from a finite from a finite location it's not just broadly across you know across the ecosystem

[00:11:40] of Google and data ingesting right so this is fit for purpose type information that you're bringing in

[00:11:47] to have them to have them test and try out the different language the sheer language difference

[00:11:53] that's right that's right I think I mean worry why right so we can't rely on the internet

[00:11:58] yeah but but I think I think I mean you probably seen this and if you haven't it's really

[00:12:03] pretty remarkable is that on the internet though and if anyone has specific subscription to

[00:12:10] the equivalents on the internet in a broad way and I know I have that on a personal basis as well

[00:12:17] and see my daughter does two inches it for university now it's kind of accepted right that we're

[00:12:23] all using that if but on the broader internet aspect of things you can also ask those questions

[00:12:30] the fact is you can't guarantee it's going to be accurate but I don't know if you know this but they've

[00:12:34] actually had the bots that are out there in the internet take like the CPA exam and other exams

[00:12:40] and they've come back and the law exams in certain countries etc they come back and actually

[00:12:45] ace all of them I don't know if you know that yeah I've seen I've seen some of those yes I have

[00:12:49] seen that as well I've seen some tests where they've laid they've had a AI and a lawyer go head to

[00:12:55] head on reviewing a very extensive and you know gnarly contract that had purposeful errors in it

[00:13:02] and the speed and the accuracy at which the AI could do it was in minutes and the lawyer took like

[00:13:06] you know five hours or something and had a couple of errors so yeah it's it's absolutely remarkable

[00:13:12] but maybe to underscore what you're talking about Julie and share you should maybe opine on this is

[00:13:17] it's all about the large language model right the data that you have to train that AI which you guys

[00:13:21] must have rich rich data from your transformation work your payroll work and not just that generic

[00:13:27] data that's anywhere right so that that stuff is very important to training your training your AI

[00:13:32] that's right and I think when it comes to payroll I mean we all know payroll right it's expected

[00:13:38] to be 100% perfect no one cares if it's perfect but certainly mistakes are not warranted or allowed

[00:13:46] and or I guess we all know the reality right sometimes they do happen but I think as a whole

[00:13:52] it's about really controlling the data and you're right that that's really secret sauce on that

[00:13:57] the other thing that you might be interested when we did this first use case and we did it not only

[00:14:03] as a pilot but then we started getting clients involved what we did as a POC I'm sorry and then

[00:14:08] we did as a pilot to get clients involved and and what we did is we actually did it under lots

[00:14:14] and lots of control right and again the second phase what we call it a crawl is without PII right

[00:14:20] because this I mean again this model and this this technology is learning and we're not we can't

[00:14:26] predict what are the answers it's going to be you know it's going to give without a lot of

[00:14:31] prompt engineering a lot of trials and so the second thing we did was we said okay here's

[00:14:36] I have hypothesis it can actually increase client sad it can decrease accuracy it can decrease time

[00:14:44] and increase efficiencies and I as you probably know the research shows that both millennials

[00:14:49] and Gen Zers they've rather chat than talk anyway right so it's kind of where the next generation

[00:14:55] of workforce is going and I can tell you not only did we see that it could answer all of those items

[00:15:01] but we saw a 30% increase in accuracy from a human agents and human you know assisted with chat

[00:15:09] bots with regular chat bots we saw a 25% customer satisfaction increase we saw the time take

[00:15:16] for average closing of tickets from a couple days or you know many many hours to less than 23 seconds

[00:15:25] and we also saw the first resolution which if as you know with employee increases like the holy

[00:15:31] grill how many of these can we get the question and the answer right away actually double so that

[00:15:37] was pretty amazing yeah that is impressive I take what I'm excited about is to see it apply to

[00:15:43] compliance I think that's going to be a tremendous helping hand right augmentation for the

[00:15:48] practitioners having to scour all that and maintain maintain systems and the opportunity for

[00:15:53] Gen I and that area is just super exciting for sure sure you foreshadowed right when you started

[00:15:59] this story about the journey and beginning with your hypotheses and the inquiry support you foreshadowed

[00:16:05] that then you know as you start to crawl you're gonna get into okay what happens when we introduce

[00:16:11] PII so I almost can't wait to hear that and I'm hoping that's where you're gonna go next

[00:16:17] sure you start to tamper with PII and not pull your hair out or sleep none at night

[00:16:24] there are so so I think that the second thing then we did and this is only a first use case

[00:16:29] obviously there's lots of other stuff going on but um is we said okay now that we've figured out

[00:16:35] that we can prompt engineer back to the regulatory compliance and really we put regulatory

[00:16:40] compliance we put anatomy of a pace up because you know no employee really understands their

[00:16:44] pace of either right and we put company policies just so you know so the first crawl part to company

[00:16:50] policies linked it with regulatory data and also linked it with anatomy of a pace up just

[00:16:56] so that we had that as a baseline and we had to add synonyms I mean we had to add a lot of other

[00:17:02] things that you would think about especially because remember we removed our functionality from

[00:17:08] the internet right we didn't want it to find things on the internet so we had to teach it

[00:17:12] with that fuel and petrol for example are at the same word um but or synonyms and mean the same thing

[00:17:18] so um so then what we did is we went to a walk phase where we added PII data and um I will tell you

[00:17:26] when we first started and we wanted to see about hallucinations which you may have heard about

[00:17:31] that are sometimes common and basically hallucination is the bot makes up stuff right

[00:17:37] and um and also we had to look at you know how what how would it if we put prompt

[00:17:43] engineering to try and tell it what it could not answer would it actually abide by this things

[00:17:49] um so we entered in PII we did it really in a very controlled environment

[00:17:54] and quite frankly as we started this the first thing we did was we entered synthetic PII if

[00:17:59] you wouldn't know the truth because we weren't ready yet to make that a huge jump to hey let's have

[00:18:04] PII in there so we added synthetic PII and we really learned a lot I mean what we learned is

[00:18:10] is a couple things one is that we had to do a lot more prompt engineering so we could actually

[00:18:14] understand the information it was getting we we did have to go back and say look you know

[00:18:21] here's the rails like for instance you can't if someone asks you hey what a surey Sullivan

[00:18:27] make and how does that compare with everybody else that that's you know there are questions

[00:18:32] that you can't answer and the thing that around this technology which is so awesome is

[00:18:36] you don't have to give it question answer question answer but on the flip side it's super risky

[00:18:42] if the fact that you can't protect a humans and what the human's going to ask it and therefore

[00:18:46] you can't protect what it necessarily will will answer so we need to make sure that we put the

[00:18:51] rails and so we had some of that but but I can tell you the learning from that is we saw pretty

[00:18:57] similar type of results that was able to ingest able to understand and able to actually control

[00:19:04] who it actually shared the information to and how so it was able to actually learn and segment

[00:19:11] what it's allowed to do with some of those rails so that was pretty interesting the other thing

[00:19:15] I mean just to mention around this we also wanted to make sure there wasn't discrimination in bias

[00:19:19] right which is something no one talks about but certainly you know and and also that there was

[00:19:25] confidentiality and data privacy aspects actually contained and respected and that was pretty

[00:19:33] challenging I'm thinking even just the the roles based access or permissions you know that

[00:19:40] I don't know if that's part of the PII but you would have had to have some you know some frame

[00:19:45] of reference for whether a manager can access information for their team or you know who who has

[00:19:51] permissions to touch what and that will that's right that's right no we had to we had to go back

[00:19:57] and say is that in order is it clear right did we think through all of the different data sources

[00:20:04] and elements and is there any difference in that and you know I mean because of what we've done

[00:20:11] and because we're EY we took a very conservative approach to that but that was really important

[00:20:16] and then I think so so we were able to see that and quite frankly we were able to now see the

[00:20:22] bot take on more than just you know you know I'm a I'm a Hungarian employee and I am going to have

[00:20:29] twins next month and I'm taking paternity leave and what are my benefits and how is that going

[00:20:35] to manifest itself in my pay that was the crawl phase it was able to answer that but now we were

[00:20:40] able to say um please give me the calculations and scenarios as to what are my options should I

[00:20:47] take paternity on the first day that my baby's born should I take it on a month later should you

[00:20:52] whatever and we saw pretty complex answers coming back with some really good calculations around

[00:20:59] the synthetic data it wasn't perfect but it actually was was much more in-depth around those type

[00:21:06] of answers which was again super impressive we tried it in many many languages um and so that's kind

[00:21:13] of that walk phase um so and then the run phase I'll just get to is really using real PII data

[00:21:21] and we've just started running into that area maybe maybe running is the wrong word it's called a

[00:21:28] run phase but we've been tiptoeing into that area with very controlled pilots to make sure

[00:21:35] that we can control them that we understand who has access back to your workflow and that we

[00:21:40] can actually test them and as you can imagine with client pilots usually a lot of that is tested

[00:21:46] with some of the individuals that are part of the the the testing team right um they're actually

[00:21:52] testing on their own data um so that they can actually give you feedback as to is the body able to

[00:21:57] do that or am I seeing other things and of course everyone in payroll so I don't know if there's

[00:22:01] any questions on that but that's kind of the the process and where we are we we have not yet

[00:22:06] published this as a feature um on our toolset um we certainly have a path toward that but as you

[00:22:13] can see we're being extremely cautious as to how do we do that and you know some of that's because

[00:22:18] this is payroll right yeah if it was something else it would be different yeah yeah no that's great

[00:22:22] that's great you you have to go slow I have a technical thought you know as you as you were talking

[00:22:27] about the um the walk phase right and so you got to a point where there's some complex calculations

[00:22:32] and some very good information that's coming out of there maybe not perfect but you know it's

[00:22:38] pretty it's pretty solid in there's some good indication so then so then what is the approach

[00:22:42] from there is it I have to get it to perfect because it's something that touches payroll or is it

[00:22:47] you know like it can be fairly good and there needs to be we need to have disclaimers and

[00:22:52] forward-looking statements you know like all of all that sort of thing I mean what's your what is

[00:22:57] the approach to refine from there or the tolerance for perfection or imperfection yeah so I think

[00:23:05] so I think the tolerance overall I mean if I'm very blind is that the the the bond actually answers

[00:23:12] uh things accurately up to 95 percent right um I think that's what we're looking at as well as

[00:23:18] working with clients um that are doing these pilots with us as to how are they going to actually

[00:23:23] what's the governance structure that they're gonna put as well on their side to actually help answer

[00:23:28] and confirm these next 5 percent and I think also it's really kind of a slow roll to begin with

[00:23:34] until we get really bullish that we have enough evidence and enough data points to know that this

[00:23:38] really works is one of the other things is to really look at um how do you have controls and the

[00:23:45] one thing we built so so I should tell you what's on the back end first of all on the back end of

[00:23:50] this we build how do we understand the thumbs up thumbs down and feedback from every single question

[00:23:56] so we can actually track how it's learning of one of the things to Julie and Pete you all think

[00:24:00] this is hysterical we had to actually train our testers not to play around with it

[00:24:08] think about think about this think about how you normally demo system you try and break it so

[00:24:12] actually even if the answer is right you put it down arrow because you want to see what it will do

[00:24:17] well with this kind of technology you can't do that because it actually thinks it's really wrong

[00:24:22] and then probably there's the potential that the next time that question is answered it will answer

[00:24:28] incorrectly increases right so you have to always think about yeah you just did a dream job you know

[00:24:34] five minutes ago break the bot you know person yeah and then you just took it away

[00:24:42] you know it's funny reminds me of I had a conversation with a a a large employer at HR tech

[00:24:48] i'm sure you've been Sherry uh and we were sitting and we were talking about change management

[00:24:52] and I said I asked it reminds me exactly of this I said what do you do for change management as

[00:24:57] you roll out these new technologies she goes you know candidly our employees somewhat lead it

[00:25:03] organically themselves and I said what do you mean she goes well we provide FAQs and and an

[00:25:08] overview like a video or something for how to do whatever the new changes she said but the reality

[00:25:12] is is they take it and play with it and they learn and then teach each other and she said so

[00:25:17] it's become like this really um amazing change management process where we have them eager to engage

[00:25:23] with it and play around and we encourage that and they ultimately learn it and pass that on so

[00:25:28] it kind of sounds the same way right like let them tinker around and get comfortable with it

[00:25:32] and learn how to use it right and break it yeah that's right that's right and and the other thing

[00:25:37] we've done Julia's as well as our P.O.C. break the bot instances are separate from the client pilots

[00:25:45] that we're doing for exactly that reason because um you know the first one was let's break the

[00:25:51] bot let's do everything we can and then the second one was oh gosh it's learning when we actually

[00:25:56] even if it gives a good answer and we do thumbs down I give it feedback when I was learning that

[00:26:02] that's not the right answer and it's searching for the right yet you know so we were like okay we

[00:26:05] we have to have pristine environments and then quite frankly each client pilot with us right now

[00:26:10] has its own environment oh I'll be very blunt with you on that um and on the back end we have so

[00:26:16] we have lots and lots of detail around what were the questions that were asked what was the answer

[00:26:21] so we do have team members actually looking through and doing the quality control check on those

[00:26:26] right now um until we are you know and measuring every single metric we can on this to make sure

[00:26:32] that we can control it um but the other thing is there's a full and there's dashboards on that

[00:26:37] there's a full list of what are those items the team looks at that but but we've also done something

[00:26:42] where within EY we've actually published this skill not the one with PII but the one without

[00:26:49] we've published the skill within our own tech as well so that it can be used by the broader EY

[00:26:55] community overall and you know we're almost 400,000 people around the globe so that we can also

[00:27:01] learn from them and from their use of the regulatory compliance knowledge and all of that so

[00:27:06] yeah that's kind of how we're going I think I think the first client that goes live you know as we

[00:27:11] go into that it will be a very measured we'll have a lot of data on those countries will roll out

[00:27:17] likely country by country or I should say country legal entity one at a time um and really do it in

[00:27:25] a safe way until we have a lot of confidence and again a lot of this is because it's payroll like

[00:27:29] we use Gen AI all the time on our personal right um with the ways we interact with different

[00:27:34] applications etc but but this is this is a real payroll application we need to make sure that

[00:27:39] it's protected oh yeah and Sherry it sounds like you you guys worked very closely with your clients

[00:27:44] which I think is fantastic with any innovation I'm sure you've learned that over the years

[00:27:48] but like how much do you are you maybe your client because you guys are dealing with some of the bigger

[00:27:53] enterprises out there right very very very savvy firms are they coming to you with this is new to

[00:27:59] us new to our organization or they coming to you saying hey we're performing pretty well in some

[00:28:03] other areas of our business maybe on the front end or in other product areas where they're already

[00:28:07] having some learnings with AI and not coming to you totally green or are you more or less um

[00:28:12] maybe the tip of the spear for them as far as AI goes well I think if you were to ask me last

[00:28:17] year when we were first doing the POCs and the first client pilots which kicked off in June of last

[00:28:22] year and I think we all knew that this technology was widely available not until January so on

[00:28:28] that we were kind of tip of the spear but I can tell you since that time what's happened and

[00:28:33] I think if you talk with any C-suite Gen AI is on their radar right how do they use it from production

[00:28:39] and how do they use it back office and how they actually lean into this and I think I think what we're

[00:28:44] seeing is that companies now have most of them have identified a Gen AI kind of leader within the

[00:28:53] organization and they're actually a ton of work cases use cases and they're working to work that

[00:28:58] and so it's more of a partnership we still are leading in many ways and in some cases we actually

[00:29:03] get just partnered up with the initiative they're doing and I think Mrs. this podcast is for

[00:29:09] you know payroll leaders I would strongly encourage that if they haven't done they actually connect

[00:29:14] within their organization and figure out who's actually leading this what are the use cases

[00:29:19] and how do they get involved I personally think when I talk with organizations that we are not

[00:29:25] doing enough and I say we in the broader sense around POCs around pilots and around use cases

[00:29:31] and showing and I'll just give an example one of the things I didn't mention about all of this

[00:29:36] first use case was just the impact on cost and actually we estimate it with clients and of course

[00:29:42] it differs per client but some of them you know that it would save about 80% of their current cost

[00:29:47] to serve clients I'm sorry serve employees on their their questions and that is significant

[00:29:54] with every company I'm dealing with they're looking for how do they actually save back office money

[00:29:59] become more efficient have a better employee experience and so if you think about that that's

[00:30:03] pretty dramatic so I think it's I think most companies are trying to figure out what to do I think

[00:30:09] unfortunately sometimes payroll leaders don't feel empowered enough to get involved and be at the

[00:30:14] tip of the sword for this and this I think about all the data and payroll and you know we can go

[00:30:19] to use case two if you'd like but really this is the time to shine it is oh it is it is they

[00:30:24] have such rich skills and now this this augmentation is just gonna be a blessing so I love it

[00:30:29] I do wonder I know I know we have folks that will want to hear some more of the use cases but I

[00:30:33] did want to ask before that so does the way that you're approaching this tend to have you focusing

[00:30:40] on broadly on use cases that will work or apply you know universally or in every country or

[00:30:48] does the approach tend to have you looking at kind of unique processes or countries and figuring out

[00:30:56] how do we optimize what's going on in there I mean what's the balance and and kind of where is

[00:31:01] this taking you on your journey yeah so I think so if you think about the different use case so for

[00:31:08] the first one that we talked about which was you know really answering employee queries and

[00:31:12] and also quite frankly administrator queries right you mentioned Pete the whole lawyer

[00:31:18] example I think if you can have from regulatory compliance perspective I think those are

[00:31:23] universal so I think those doesn't matter which country where you're located what language you speak

[00:31:28] those use cases are universal and the pain points are universal right I think if you think about some

[00:31:33] of the other use cases so some of the things we're doing with clients is we actually have done some

[00:31:38] digital worker type of projects and if you don't know what digital worker is it's actually for

[00:31:46] better reason it's a low code RPA so it functions just like you would an RPA but now it's low code

[00:31:52] because it's LLM and it also can be easily programmed and used by business leaders and by individuals

[00:32:01] in the business so if you think about all these manual processes that even though we've kind of

[00:32:06] moved from one stage to the next they still play payroll what you find is you find that before

[00:32:13] you had to have RPA only worked if it was highly repeatable you had to have somebody specific with

[00:32:19] the programming knowledge and language to get in and actually do it and now you don't have to do

[00:32:24] that anymore with things like the digital worker you can say we have a use case with a client I'll

[00:32:30] use just one of them where they have in Switzerland they have a ton of child allowances it's a huge

[00:32:36] process there's no program really developed to handle everything they have to handle including

[00:32:41] policy checks approvals all of those things and it's just taking a ton of time for the payroll team

[00:32:47] to actually get ready for because as you may know payroll teams tend to have to do a lot of these

[00:32:53] things and so they were able to actually very quickly actually take this low code RPA put it

[00:32:59] together and and actually it's saved you know hundreds of hours every single month just by doing that

[00:33:07] and so I think to your question Julie I think those type of technologies now allow you to really

[00:33:12] look specifically at the country process oh yeah low code and no code microservices has been huge

[00:33:18] I think for payroll not having to involve IT to go write a report or make some sort of change to a

[00:33:23] table or a pay code it's such a blessing the speed and agility of that it's just amazing

[00:33:30] yeah yeah so we also have some other use case that I don't know if you're interested in but

[00:33:35] what yeah we're one or two fears this would be great yeah sure sure so one is around you know reporting

[00:33:44] right so certainly what we've been able to do is to see this new technology really focus on

[00:33:52] reporting and it can be as simple as something like a copilot and I don't know how many of you have

[00:33:58] copilot on your you know if you're a Microsoft shop on your your machines but that's been pretty

[00:34:04] amazing because we know payroll teams do lots of excel jockey and now the with this technology

[00:34:10] it's able to do it for you so you can actually say you know compare these three files and tell me

[00:34:14] the anomalies for example and that comes right with the the feature you just have to learn as a

[00:34:20] user how to use it another one is around you know broader reports right and and now there's

[00:34:27] the technologies which allow again normal language to actually go into your data model you've got

[00:34:32] to have your data model in order right but go into your data model and actually answer questions

[00:34:38] and and so long gone are now the times when I have a 41 different reports because someone might

[00:34:44] ask me for it and I'm afraid I won't have it and I'll be cut off guard but now is the time where

[00:34:49] you can actually say look I have the data model is put in the same database it's it's mapped correctly

[00:34:56] and therefore I can rely on the technology sitting on top to actually ask it questions in

[00:35:01] normal language and it can immediately produce the results give me a chart give me something

[00:35:06] because we love Excel to export out an excel etc so that's really changing the workload I think

[00:35:12] when I talk with payroll teams they spend tons of time on the source to grow tens of times on the

[00:35:17] control right and tons of times on the reporting at the back end and so that's another use case

[00:35:23] that we're seeing and and that's not just such a use case it's actually out in the market and being

[00:35:27] used so I don't know if there's any questions on that from an analytics perspective I've been thinking

[00:35:33] about the fact that I've probably spent the better part of my career with a goal to actually become

[00:35:39] a whiz at pivot tables and I spent my entire career not becoming a whiz at pivot tables or

[00:35:47] access databases to fill gaps yeah yeah I can actually fill some of my own gaps I hope in

[00:35:54] you know the next several years and I can actually spend a lifetime never having to be

[00:35:59] the the access databases Sherry you'll appreciate this Julie how many times have you gone into an

[00:36:03] organization for like a due diligence and you find like oh we and this is years ago probably more now

[00:36:08] now than now is that you'd have some you know some guy Kevin or whatever down and you know

[00:36:13] down in the basement and he maintains these three access databases that hold payroll together

[00:36:18] and without him we can't we can process payroll and I'm like what are you talking about

[00:36:22] that's microservices right that's integrations you can now bring a solution to that to that problem

[00:36:27] versus having to go build something so yeah pivot tables and access databases power payroll boy

[00:36:33] they power more payroll than you would think that's right they do they do and I think I think

[00:36:38] with this new technology as we move into it it's we're still going to need good database right

[00:36:44] with all the data we're going to need but but some of the one of the other use cases I didn't

[00:36:49] mention is this whole payroll mapping so you have a database and that's great but if you don't

[00:36:55] have a data model then it's consistent and it could be for one country for two for three for four

[00:37:01] but but now what we're you're able to do is use this technology and this generative AI to actually

[00:37:06] do the mapping for you yeah which is pretty cool and also have the controls around it right because

[00:37:13] we've all seen mapping that's been done where you're supposed to have I don't know 200 different

[00:37:18] items and you end up a 2000 because everybody thinks that this thing is special right so now

[00:37:23] you also have some governance around it as well and that's certainly a use case that you know

[00:37:29] that we're applying but we're also seeing clients apply as well to these databases that they've

[00:37:34] actually created yeah are you seeing that share is kind of the um an alternative to the whole

[00:37:41] integration bit or even implementation right that's where I tend to hear a lot of excitement

[00:37:46] around that area is data migration all the data mapping or integration that has to happen as

[00:37:52] opposed to the ongoing care and feeding of that that seems to be a you know a peak interest is

[00:37:58] in that one time burst of effort is that also how you're seeing some of that applied or is it as

[00:38:02] much ongoing as but well right now it's really ongoing but let's talk about implementation

[00:38:08] with this new technology you can actually be in a position where it actually pulls the data all

[00:38:14] you need is the big data down it it should be able to actually compare it to what you need to do

[00:38:20] business in that country it should be able to actually look and see what are those things that

[00:38:27] are outside of it and actually help you really clean up the data so instead of I don't know about

[00:38:33] the implementation or requirement gathering process is got to be one of the most

[00:38:40] difficult ones I'm being nice here um processes in payroll and my hats off to all those people

[00:38:46] and implementation that works so hard on this but this technology should change their job

[00:38:51] dramatically I haven't seen it yet but I can see it coming and in that you'll be able to take just

[00:38:57] a big data dump of you know last 24 months that should be able to compare it to what is it

[00:39:03] that you need in the country along with what are the regulatory requirements and be able to spit out

[00:39:07] here's the exceptions that need you need to dig in with your client and that would be an amazing

[00:39:12] amazing change in implementation. Hey what payroll person out there isn't thinking parallel testing

[00:39:19] days oh oh oh yeah or it'll take minutes yeah full different thing right minutes to do

[00:39:27] parrots versus you know once yeah no it's beautiful it's a beautiful thing awesome well

[00:39:32] it's so helpful to get some specific use cases are there for payroll share here there are

[00:39:37] there's touch points that jump out at you more than others like I know we've talked about some of

[00:39:42] the front end ones and starting with the increased support and and maybe modeling stat you know changes

[00:39:47] that drive what payroll is going to look like but anything in time and attendance or GL that's

[00:39:53] that's jumped out at the top of the list or other touch points that are just painful leaves for

[00:39:58] for payroll. Yeah well I think I think um well so one of the most difficult things that I think clients

[00:40:05] deal with our leave management right and now with this type of technology I think we're going to see

[00:40:12] just a real difference in the ability to actually manage those around the globe and because as

[00:40:18] you know they're different for every single country they're pretty complex depending on which

[00:40:23] country you're involved and I know there's lots of systems out there but the ones that hand to

[00:40:28] leave management well are the local to local ones um and and I think that that's coming I haven't

[00:40:35] seen a product yet around that but I can feel that that is probably the next in the big area

[00:40:41] and it kind of relates to this digital worker use case that I mentioned before in that it's really

[00:40:47] country specific a big big issue um and lots of requirements around that whether it's reporting

[00:40:54] to the government whether it's affects actually the net or gross pay you know all of those things so

[00:40:59] I think that that is probably one um key area that we're going to see I think as far as GLs we

[00:41:06] just talked about the payroll mapping um I mean I think from a GL perspective yeah absolutely I

[00:41:12] think the ability for this technology to actually not follow that mapping all the way through to

[00:41:17] geocreation and also balancing and again imagine if you work in a payroll team and you actually

[00:41:24] have this technology in order then when it does it actually spits out to not that the GL file

[00:41:29] actually didn't go through but what were the area areas before it tried to send

[00:41:34] that don't balance that need to be looked at and confirmed by the human nature and I think

[00:41:38] the last thing I would just say is one of our approaches is human at the center for all of this right

[00:41:44] and really I think in order our key learning around gender to the eye has been that people need

[00:41:51] to be at the center of the transformation agenda and drive it and when they are and they need

[00:41:57] to create the vision and when they are I think what we've seen is that you have about a two and a

[00:42:02] half times acceleration in the rate of transformation success and so that's really important. I do

[00:42:10] think all of us have to re-skill you heard me say at the beginning you know I'm an old dog and so

[00:42:15] I'm able to learn some new tricks which is pretty amazing but um I think you know re-skilling

[00:42:20] through the future and that really goes to how do I connect you know how do I take classes I

[00:42:25] actually working on an AI badge how do I take classes how do I understand as well as quite frankly

[00:42:32] I have someone on our team who is totally into this he's really amazing I call him my AI Yoda

[00:42:38] and I go to him for absolutely everything and so I think all of us in payroll I think the skills

[00:42:43] we have today and the skills we need tomorrow are completely different. Awesome you know I have one

[00:42:49] list question sharing because I know we've been at this for a little bit but I feel like we would

[00:42:53] be missing something if we didn't ask you about business case and how are payroll leaders that you're

[00:43:00] interacting with today moving faster or creating you know the case for change or figuring out how to

[00:43:08] actually get a little bit of traction in this space. What are there tips and tricks are there

[00:43:14] things that you're seeing that successful payroll leaders are using to actually dig in and

[00:43:20] and make something happen here? Yeah and are you moving fast enough right I don't I

[00:43:24] debate that their that organizations firms I'm practitioners I'm talking to they don't feel

[00:43:28] their organizations are are are reacting quickly enough. I think I mean you're right when everyone

[00:43:33] is focused on hey I need to take all this journal to the AI I need to actually put it in the front

[00:43:39] part of my product so I can actually sustain and live as a business what we're seeing is any back

[00:43:44] office function is taking a back seat right from a transformation perspective and payrolls no

[00:43:49] different I think and also the company may not figure out like how do I have the funds to make

[00:43:54] the changes in the business case that I need. I think and you all know when you look at

[00:43:59] traditional payroll business cases there's actually tangible costs and there's a ton of

[00:44:03] I intangible I'm always reminded of the iceberg example but those intangibles the cost is out

[00:44:10] there it's just about how do you actually quantify and pull it together so what what I'm seeing

[00:44:14] as a couple things one which you know I don't tend to be a risk first person but one is a lot

[00:44:23] of my clients are having a lot of success by really putting out the risk and I'll give you an

[00:44:27] example since you know cyber attacks are certainly up there's certainly up on payroll there's a

[00:44:33] lot of companies out there that have managed to do payroll cheap and cheerful but with lots of

[00:44:38] risk around their technology around their infrastructure around their backup that is certainly

[00:44:44] several of the clients I deal with are really taking a risk based approach and showing and proving

[00:44:50] that money is being spent anyway and what they're doing is kind of penny wise and pound full

[00:44:54] wish if you will as well as increasing the risk so that's one approach I think that what I'm seeing

[00:45:00] a lot of leaders do is really figure out how do I elevate payroll within the C suite and within my

[00:45:06] company and then how do I actually identify quantify as part of my business case all of the leakage

[00:45:13] or items that are happening around payroll and actually how do I actually set up a self-funding

[00:45:18] mechanism so as those are identified I can actually share with my operation with my stakeholders

[00:45:24] in my operation leader with my sage or oh or head of VP company then with my you know a CFO

[00:45:32] so that I actually have able to self-fund my transformation through the transformation of the

[00:45:37] payroll and also the upstream processes so I think that's what I'm really seeing grab hold

[00:45:42] so I don't know if you've seen other things but those are the two kind of approaches that I'm

[00:45:46] seeing that are being extremely successful yeah cool tactics well we're at the we know we're

[00:45:52] early on right especially folks are trying to do things themselves versus as part of a large

[00:45:56] organization that does this with multiple clients so I'm sure I'm sure what's the all sorts of

[00:46:01] things here as the year unfolds great it's gonna be exciting yes it is it is it's very exciting

[00:46:07] little scary right but it's also incredibly exciting for so many things right I mean humanity in

[00:46:12] general but we have to be careful like you said decoshes walks you know crawl walk run protect the

[00:46:18] data protect the people and I think we can do some good stuff with it yeah share we don't want to

[00:46:23] leave today without asking where folks can reach you and what's up for yeah they are sure well

[00:46:30] I love Pera as I said before certainly anyone who's interested can reach out to me and connect

[00:46:36] with me on LinkedIn or can shoot me over an email and my emails really simple it's

[00:46:43] sherry dot Sullivan at eY dot com perfect yeah we'll list that in the description to make sure people

[00:46:48] can can get in touch with you awesome yeah thanks for coming thank you for having me yeah

[00:46:55] yeah Julie what do you have going on anything I'm on the verge of having one of those microsurvey's

[00:47:00] in HR AI that will be ready for prime time and we'll be looking to get some practitioner input on

[00:47:06] that yeah beautiful I love it yeah let me know if I can help well yeah I got a bunch of events

[00:47:11] coming up so I'll be out there in quite a few places speaking at the capital summit with the

[00:47:16] payroll org I'm doing the IPPA event out in Vegas with team ice all so yeah it's gonna be fun

[00:47:22] and of course the payroll profession competent survey is still running if you haven't gotten

[00:47:26] in on that get your voice heard we need every payroll practitioner to to to share in that so please

[00:47:31] please take that as soon as you can I really appreciate it perfect not so yeah jump in and do it

[00:47:38] I know I know well look this has been great sherry thank you so much for joining us and Julie

[00:47:41] it's always always a pleasure as well see him here until next time take care everyone